Arc Fault Detection Using Machine Learning

1. A circuit interrupter comprising: a housing configured to be installed in at least one of an electrical wall box or a load center; a conductive path; a switch configured to selectively interrupt the conductive path; a sensor in electrical communication with the conductive path and configured to measure an electrical characteristic of the conductive path, the sensor configured to provide a plurality of sensor measurements; a memory having an arc detection program stored therein, the arc detection program implementing a machine learning model, the arc detection program including a field-updatable program portion configured to be field-updatable and a non-field-updatable program portion, the field-updatable program portion including a plurality of machine learning model parameters, wherein the non-field-updatable program portion is configured to use the plurality of machine learning model parameters to decide between presence of an arc event or absence of an arc event; and a controller positioned within the housing and coupled to the memory, wherein the arc detection program, when executed by the controller, causes the controller to perform an operation comprising: computing input data for the machine learning model based on the plurality of sensor measurements, providing a decision between presence of an arc event or absence of an arc event, the decision based on the input data, and causing the switch to interrupt the conductive path when the decision indicates presence of an arc event.

2. The circuit interrupter of claim 1, further comprising a communication device, wherein the controller is further configured to cause the communication device to communicate, with a remote device, at least one of: the plurality of sensor measurements or the input data computed based on the plurality of sensor measurements.

3. The circuit interrupter of claim 2, further comprising a user-accessible actuator, wherein the controller is configured to cause the communication device to communicate with the remote device when the user-accessible actuator is actuated.

4. The circuit interrupter of claim 2, wherein the memory is further configured to store a firmware update program which, when executed by the controller, causes the controller to perform a further operation comprising: causing the communication device to receive an updated program portion having reduced false positive arc event decisions; and replacing the field-updatable program portion of the arc detection program with the updated program portion.

5. The circuit interrupter of claim 1, wherein the arc detection program implements a machine learning model having nodes, wherein the plurality of program parameters of the field-updatable program portion are node parameters of the nodes of the machine learning model, and wherein the non-field-updatable program portion is configured to perform node operations for the nodes of the machine learning model.

6. The circuit interrupter of claim 5, wherein the machine learning model is a decision tree, wherein the node parameters include, for each node of the decision tree, one of: a branch criterion and destination nodes, or a decision tree decision, and wherein the node operations comprise, for each node of the decision tree: deciding whether the node parameters for the node include a decision tree decision or a branch criterion, in the case the node parameters for the node include of a decision tree decision, providing the decision tree decision, and in the case the node parameters for the node include a branch criterion and destination nodes, apply the branch criterion to the input data to select one of the destination nodes, and repeat the node operations for the selected destination node.

7. The circuit interrupter of claim 1, wherein the plurality of sensor measurements include high frequency component measurements relating to a high frequency characteristic of the conductive path, wherein the input data includes high frequency component input data computed based on the high frequency component measurements, the high frequency component input data including at least one of: (i) a close to low count computed as: a number of the high frequency component measurements in a first low measurement range when an average of the high frequency component measurements is in a first range, or a number of the high frequency component measurements in a second low measurement range when the average of the high frequency component measurements is in a second range, (ii) a close to high count computed as: a number of the high frequency component measurements in a first high measurement range when the average of the high frequency component measurements is in the first range, or a number of the high frequency component measurements in a second high measurement range when the average of the high frequency component measurements is in the second range, (iii) an index difference between a first index corresponding to a maximum increase between consecutive measurements of the high frequency component measurements and a second index corresponding to a maximum decrease between consecutive measurements of the high frequency component measurements, (iv) a low count computed as a number of the high frequency component measurements in a third low measurement range, wherein the third low measurement range is below the second range, (v) a first high count computed as a number of the high frequency component measurements in a third high measurement range, or (vi) a second high count computed as a number of the high frequency component samples in a fourth high measurement range.

8. A method in a circuit interrupter having a conductive path, a switch configured to selectively interrupt the conductive path, and at least one sensor in electrical communication with the conductive path and configured to measure at least one electrical characteristic of the conductive path to provide a plurality of sensor measurements, the method comprising: executing an arc detection program in the circuit interrupter, the arc detection program implementing a machine learning model and including a field-updatable program portion configured to be field-updatable and a non-field-updatable program portion, the field-updatable program portion including a plurality of machine learning model parameters to be used by the non-field-updatable program portion for deciding between presence of an arc event or absence of an arc event, wherein executing the arc detection program causes the circuit interrupter to perform an operation comprising: computing input data for the machine learning model based on the plurality of sensor measurements, deciding between presence of an arc event or absence of an arc event, based on the input data, to provide a decision, and causing the switch to interrupt the conductive path when the decision indicates presence of an arc event.

9. The method of claim 8, further comprising communicating, with a remote device, at least one of: the plurality of sensor measurements or the input data computed based on the plurality of sensor measurements, when the decision indicating presence of an arc event corresponds to a false positive arc event decision.

10. The method of claim 9, further comprising: receiving an updated program portion having reduced false positive arc event decisions; and replacing the field-updatable program portion of the arc detection program with the updated program portion.

11. The method of claim 8, wherein the arc detection program implements a machine learning model having nodes, wherein the plurality of program parameters of the field-updatable program portion are node parameters of the nodes of the machine learning model, and wherein the non-field-updatable program portion is configured to perform node operations for the nodes of the machine learning model.

12. The method of claim 11, wherein the machine learning model is a decision tree, wherein the node parameters include, for each node of the decision tree, one of: a branch criterion and destination nodes, or a decision tree decision, and wherein the node operations comprise, for each node of the decision tree: deciding whether the node parameters for the node include a decision tree decision or a branch criterion, in the case the node parameters for the node include of a decision tree decision, providing the decision tree decision, and in the case the node parameters for the node include a branch criterion and destination nodes, apply the branch criterion to the input data to select one of the destination nodes, and repeat the node operations for the selected destination node.

13. The method of claim 8, wherein the plurality of sensor measurements include high frequency component measurements relating to a high frequency characteristic of the conductive path, wherein the input data includes high frequency component input data computed based on the high frequency component measurements, the high frequency component input data including at least one of: (i) a close to low count computed as: a number of the high frequency component measurements in a first low measurement range when an average of the high frequency component measurements is in a first range, or a number of the high frequency component measurements in a second low measurement range when the average of the high frequency component measurements is in a second range, (ii) a close to high count computed as: a number of the high frequency component measurements in a first high measurement range when the average of the high frequency component measurements is in the first range, or a number of the high frequency component measurements in a second high measurement range when the average of the high frequency component measurements is in the second range, (iii) an index difference between a first index corresponding to a maximum increase between consecutive measurements of the high frequency component measurements and a second index corresponding to a maximum decrease between consecutive measurements of the high frequency component measurements, (iv) a low count computed as a number of the high frequency component measurements in a third low measurement range, wherein the third low measurement range is below the second range, (v) a first high count computed as a number of the high frequency component measurements in a third high measurement range, or (vi) a second high count computed as a number of the high frequency component samples in a fourth high measurement range.

14. A processor-implemented method for updating installed arc fault circuit interrupt (AFCI) devices configured to execute an arc detection program implementing a machine learning model, the arc detection program including a field-updatable program portion configured to be field-updatable and a non-field-updatable program portion, the field-updatable program portion including a plurality of machine learning model parameters to be used by the non-field-updatable program portion for deciding between presence of an arc event or absence of an arc event, the method comprising: receiving, from the installed AFCI devices, event data including at least one of: (i) a plurality of sensor measurements provided by sensors, in the installed AFCI devices, configured to measure at least one electrical characteristic of conductive paths in the AFCI devices, or (ii) input data for the machine learning model computed, in the installed AFCI devices, based on the plurality of sensor measurements; preparing, based on the event data, an updated program portion having reduced false positive arc event decisions; and communicating, to the installed AFCI devices, the updated program portion as a replacement for the field-updatable program portion of the arc detection program.

15. The processor-implemented method of claim 14, wherein preparing the updated program portion having reduced false positive arc event decisions includes: in case the event data is the plurality of sensor measurements, computing the input data for the machine learning model based on the plurality of sensor measurements; associating the input data with a label indicating absence of an arc event; training an updated machine learning model using the input data and the label; and providing parameters of the updated machine learning model as the updated program portion.

16. The processor-implemented method of claim 14, wherein the arc detection program implements a machine learning model having nodes, wherein the plurality of program parameters of the field-updatable program portion are node parameters of the nodes of the machine learning model, and wherein the non-field-updatable program portion is configured to perform node operations for the nodes of the machine learning model.

17. The processor-implemented method of claim 16, wherein the machine learning model is a decision tree, wherein the node parameters include, for each node of the decision tree, one of: a branch criterion and destination nodes, or a decision tree decision, and wherein the node operations comprise, for each node of the decision tree: deciding whether the node parameters for the node include a decision tree decision or a branch criterion, in the case the node parameters for the node include of a decision tree decision, providing the decision tree decision, and in the case the node parameters for the node include a branch criterion and destination nodes, apply the branch criterion to the input data to select one of the destination nodes, and repeat the node operations for the selected destination node.

18. A system for updating installed arc fault circuit interrupt (AFCI) devices configured to execute an arc detection program implementing a machine learning model, the arc detection program including a field-updatable program portion configured to be field-updatable and a non-field-updatable program portion, the field-updatable program portion including a plurality of machine learning model parameters to be used by the non-field-updatable program portion for deciding between presence of an arc event or absence of an arc event, the system comprising: at least one processor; and one or more memory storing instructions which, when executed by the at least one processor, cause the system to perform an operation comprising: receiving, from the installed AFCI devices, event data including at least one of: (i) a plurality of sensor measurements provided by sensors, in the installed AFCI devices, configured to measure at least one electrical characteristic conductive paths in the AFCI devices, or (ii) input data for the machine learning model computed, in the installed AFCI devices, based on the plurality of sensor measurements; preparing, based on the event data, an updated program portion having reduced false positive arc event decisions; and communicating, to the installed AFCI devices, the updated program portion as a replacement for the field-updatable program portion of the arc detection program.

19. The system of claim 18, wherein in preparing the updated program portion having reduced false positive arc event decisions, the instructions, when executed by the at least one processor, cause the system to: in case the event data is the plurality of sensor measurements, compute the input data for the machine learning model based on the plurality of sensor measurements; associate the input data with a label indicating absence of an arc event; train an updated machine learning model using the input data and the label; and provide parameters of the updated machine learning model as the updated program portion.

20. The system of claim 18, wherein the arc detection program implements a machine learning model having nodes, wherein the plurality of program parameters of the field-updatable program portion are node parameters of the nodes of the machine learning model, and wherein the non-field-updatable program portion is configured to perform node operations for the nodes of the machine learning model.

21. The system of claim 20, wherein the machine learning model is a decision tree, wherein the node parameters include, for each node of the decision tree, one of: a branch criterion and destination nodes, or a decision tree decision, and wherein the node operations comprise, for each node of the decision tree: deciding whether the node parameters for the node include a decision tree decision or a branch criterion, in the case the node parameters for the node include of a decision tree decision, providing the decision tree decision, and in the case the node parameters for the node include a branch criterion and destination nodes, apply the branch criterion to the input data to select one of the destination nodes, and repeat the node operations for the selected destination node.